WEBVTT 00:00:00.020 --> 00:00:01.500 How many cores does your machine have? 00:00:02.080 --> 00:00:02.220 10? 00:00:02.720 --> 00:00:02.960 18? 00:00:03.740 --> 00:00:06.840 Well, your async Python code uses just one of them. 00:00:07.580 --> 00:00:09.060 That isn't a bug in asyncio. 00:00:09.420 --> 00:00:10.180 That's the design. 00:00:10.840 --> 00:00:16.940 And things like optimizing the event loops to be faster by maybe 20% with uv doesn't fundamentally 00:00:17.090 --> 00:00:17.480 change that. 00:00:17.760 --> 00:00:20.520 That's why Giovanni Bariliari started over. 00:00:21.180 --> 00:00:26.400 Joe is the creator of Granian, the Rust-based server that powers Talk Python and all the 00:00:26.450 --> 00:00:27.740 other things we run here. 00:00:28.020 --> 00:00:33.580 His new project is Tone.io, an async runtime written from scratch specifically for free-threaded Python. 00:00:34.280 --> 00:00:38.320 Real threads, a handful of primitives instead of async.io's pile of them, 00:00:38.820 --> 00:00:43.260 and it flat out refuses to even start if the GIL is present. 00:00:43.880 --> 00:00:49.600 This is Talk Python To Me, episode 561, recorded September 1st, 2026. 00:01:10.620 --> 00:01:14.940 Welcome to Talk Python To Me, the number one Python podcast for developers and data scientists. 00:01:15.460 --> 00:01:16.840 This is your host, Michael Kennedy. 00:01:17.220 --> 00:01:20.760 I'm a PSF fellow who's been coding for over 25 years. 00:01:21.400 --> 00:01:22.520 Let's connect on social media. 00:01:22.840 --> 00:01:26.000 You'll find me and Talk Python on Mastodon, BlueSky, and X. 00:01:26.190 --> 00:01:28.120 The social links are all in your show notes. 00:01:28.880 --> 00:01:32.400 You can find over 10 years of past episodes at talkpython.fm. 00:01:32.580 --> 00:01:35.780 And if you want to be part of the show, you can join our recording live streams. 00:01:36.120 --> 00:01:36.600 That's right. 00:01:36.790 --> 00:01:40.060 We live stream the raw uncut version of each episode on YouTube. 00:01:40.460 --> 00:01:45.060 Just visit talkpython.fm/youtube to see the schedule of upcoming events. 00:01:45.290 --> 00:01:48.940 Be sure to subscribe there and press the bell so you'll get notified anytime we're recording. 00:01:49.840 --> 00:01:52.380 This episode is sponsored by Sentry's Seer. 00:01:52.660 --> 00:01:55.020 If you're tired of debugging in the dark, give Seer a try. 00:01:55.680 --> 00:01:59.320 There are plenty of AI tools that help you write code, but Sentry's Seer is built to 00:01:59.320 --> 00:02:00.900 help you fix it when it breaks. 00:02:01.480 --> 00:02:07.200 Visit talkpython.fm/sentry and use the code Talk Python26, all one word, no spaces, 00:02:07.620 --> 00:02:09.340 for $100 in Sentry credits. 00:02:09.979 --> 00:02:11.720 And it's brought to you by us. 00:02:12.360 --> 00:02:15.440 Talk Python and Python Bytes both now have MCP servers. 00:02:16.040 --> 00:02:20.440 Point your AI at 10 plus years of Python episodes, transcripts, and show notes. 00:02:20.840 --> 00:02:25.160 Free. Click MCP in the nav at talkpython.fm and at Python Bytes. 00:02:25.760 --> 00:02:29.020 Hello, hello. Welcome back to Talk Python and me. 00:02:29.350 --> 00:02:30.020 Great to have you here, Joe. 00:02:30.280 --> 00:02:33.640 Hey, thanks for having me. It's nice to get back to the show. 00:02:33.930 --> 00:02:34.540 It's been a while. 00:02:35.040 --> 00:02:35.920 It has been a while. 00:02:36.220 --> 00:02:41.800 And we talked about Granian last time, the Rust-based web application server for Python. 00:02:42.200 --> 00:02:44.200 And I got to just take a moment and say thank you. 00:02:44.320 --> 00:02:45.000 We'll come back to this. 00:02:45.040 --> 00:02:49.400 But thanks for powering Talk Python and all the different things. 00:02:49.480 --> 00:02:52.680 I think across our web apps, I don't know, like 10 million requests a month. 00:02:52.780 --> 00:02:53.740 And it's been flawless. 00:02:54.920 --> 00:02:55.720 So thanks. 00:02:55.880 --> 00:02:58.420 I appreciate you serving up all the stuff for everyone. 00:02:58.680 --> 00:02:59.120 Oh, yeah. 00:02:59.340 --> 00:02:59.440 Thanks. 00:03:00.220 --> 00:03:00.820 It's been a pleasure. 00:03:00.980 --> 00:03:04.540 Like, it's nice to see, you know, when something just works. 00:03:04.980 --> 00:03:09.660 like when you do something and like at the end of the day it just works and people start using it 00:03:09.740 --> 00:03:16.560 like of course like they they ask for a bunch of features you know um but it's nice to see yeah it 00:03:16.580 --> 00:03:21.300 is really nice to see and i think it's i think it's good now it's been a while since you've been on the 00:03:21.400 --> 00:03:26.440 show a couple years i don't i didn't check the exact date when i look back maybe people are new 00:03:26.480 --> 00:03:30.419 to the show or haven't heard your prior episode or just don't remember like give us a quick 00:03:30.440 --> 00:03:37.160 introduction who you are yeah sure so i'm giovanni but everybody can call me joe i'm italian but i 00:03:37.160 --> 00:03:44.400 live in vienna i am working as a programmer in century but i'm a physicist so i think like 00:03:44.760 --> 00:03:53.360 that's pretty peculiar i've been like in the open source ecosystem uh mainly in python um probably 00:03:53.440 --> 00:04:00.560 for the last oh god I think like it's almost 20 years now um and fun fact like I actually started 00:04:00.800 --> 00:04:07.160 like um doing open source or or um contributing to open source projects even before actually 00:04:07.440 --> 00:04:16.900 working as a programmer so um yeah that's me I maintain like too many packages nowadays for like 00:04:16.959 --> 00:04:18.380 the energy I have. 00:04:20.080 --> 00:04:22.019 But yeah, that's pretty much. 00:04:22.360 --> 00:04:22.640 Excellent. 00:04:22.890 --> 00:04:25.320 Yeah, I know how that goes with the packages. 00:04:25.660 --> 00:04:26.740 It's like, oh, here's a cool idea. 00:04:26.740 --> 00:04:27.500 I want to put it out there. 00:04:27.800 --> 00:04:29.520 If it gains no traction, you can shut it down. 00:04:30.080 --> 00:04:32.000 If it gains a ton of traction, you're excited. 00:04:32.100 --> 00:04:33.280 If it gains a little traction, 00:04:33.780 --> 00:04:35.140 but enough, then you got to keep working on it. 00:04:35.200 --> 00:04:36.740 That's kind of not ideal, right? 00:04:37.080 --> 00:04:37.320 Yeah. 00:04:37.430 --> 00:04:39.060 And especially like nowadays with AI, 00:04:39.560 --> 00:04:41.440 like the amount of things you, 00:04:41.570 --> 00:04:42.520 oh, I have an idea. 00:04:43.740 --> 00:04:48.680 Like the chances like to just throw a bunch of agents into that and see what happens. 00:04:49.200 --> 00:04:49.840 It's very dangerous. 00:04:50.240 --> 00:04:50.820 It's very dangerous. 00:04:51.240 --> 00:04:52.040 I have an idea. 00:04:52.040 --> 00:04:52.840 It'll probably take me an hour. 00:04:52.880 --> 00:04:54.080 Let me see if I can validate it. 00:04:54.520 --> 00:04:57.080 Thursday afternoon, you're still like, oh, I'm still almost there. 00:04:57.120 --> 00:04:58.980 It's like, where did my week go? 00:04:59.280 --> 00:04:59.500 What happened? 00:05:00.360 --> 00:05:01.420 It's fun though, isn't it? 00:05:01.500 --> 00:05:02.360 It's a wild time. 00:05:02.720 --> 00:05:05.000 Yeah, it's weird, right? 00:05:05.200 --> 00:05:12.800 Because in a way, we have this very powerful, not super constant tool. 00:05:14.080 --> 00:05:19.760 but it's very powerful and and sometimes just amazing like sometimes like what do you what 00:05:19.760 --> 00:05:27.160 you get back is just amazing and you say like oh okay that's nice that's unexpected sometimes is 00:05:27.280 --> 00:05:34.960 the dumbest thing possible you can you you can get but i guess like the point the main point is that 00:05:35.240 --> 00:05:42.860 i think it it shows up how as humans we have like a complicated relationship with tools 00:05:43.860 --> 00:05:51.980 it can exploit a bunch of weird human behavior sometimes so yeah yeah if you like building things 00:05:52.520 --> 00:05:56.560 you know create you've got ideas and you want to see them come to life it's an unprecedented time 00:05:56.720 --> 00:06:03.220 if if you really your main joy was to be in the code working on the little bits of nuance i can 00:06:03.300 --> 00:06:07.300 see that it's frustrating and uh there's there's a whole there's a whole thing that we go into there 00:06:07.260 --> 00:06:08.120 but that's probably not. 00:06:08.130 --> 00:06:09.660 But wow, what a crazy time, right? 00:06:09.900 --> 00:06:10.900 Yeah, absolutely. 00:06:11.400 --> 00:06:14.000 Yeah, so let's jump into the topics here. 00:06:14.010 --> 00:06:18.560 I think the first thing I want to do is let's just do a little dive into Granian. 00:06:19.320 --> 00:06:21.520 That's why I reached out to you to have you on the show the first time. 00:06:21.840 --> 00:06:23.160 Like I said, it's powering. 00:06:23.390 --> 00:06:25.460 I have on the Talk Python server, 00:06:25.680 --> 00:06:29.680 there's 33 different containers running a bunch of different things, 00:06:29.770 --> 00:06:32.440 and probably about 15 of them are running Granian. 00:06:33.020 --> 00:06:33.380 Nice. 00:06:33.880 --> 00:06:34.160 Yeah. 00:06:34.800 --> 00:06:36.820 So like I said, it's been really, really cool. 00:06:37.040 --> 00:06:41.400 just tell people what is Granian because I think not only is it interesting as a thing that you've 00:06:41.400 --> 00:06:45.580 done previously and been on the show before it's also an interesting lead-in to where we're going 00:06:45.580 --> 00:06:52.940 with Tonio so what's great yeah sure uh so Granian is an application server for Python so in Python 00:06:53.540 --> 00:07:02.700 for historical reasons um we separated like the application layer from the HTTP layer so when we 00:07:02.600 --> 00:07:07.640 want to serve a web application in Python compared to other languages 00:07:07.740 --> 00:07:12.260 in which the server part is actually a part of the application. 00:07:12.640 --> 00:07:18.640 In Python, we tend to have a separated package, which is a server, 00:07:19.100 --> 00:07:22.340 and that server's import your actual application. 00:07:22.760 --> 00:07:25.820 And then we have application protocols 00:07:26.720 --> 00:07:30.120 to make the server talk with your application and back. 00:07:31.100 --> 00:07:34.960 The two main protocols in Python are VUSGI or WSGI. 00:07:35.500 --> 00:07:38.120 I never really understood how to say that, 00:07:39.720 --> 00:07:44.760 which is like the original PET333, if I'm not mistaken. 00:07:45.050 --> 00:07:48.100 So it's quite ancient in the Python history. 00:07:49.140 --> 00:07:53.860 And once we had like AsyncIO, a new protocol emerged. 00:07:54.380 --> 00:07:57.660 That's called ASGI or ASGI because, of course, 00:07:57.780 --> 00:08:00.080 it's the asynchronous server gateway. 00:08:00.520 --> 00:08:05.260 So, Granian is, again, an application server, 00:08:06.360 --> 00:08:11.720 so you can consider it as an alternative to MicroVoosgi, 00:08:13.860 --> 00:08:17.600 Uvicorn, Hypercorn, and a bunch of other servers 00:08:17.820 --> 00:08:19.240 that happened during the years. 00:08:20.520 --> 00:08:24.640 And, yeah, the peculiar fact is it's made with Rust. 00:08:26.080 --> 00:08:29.880 So, compared to all the other servers you might use in Python, 00:08:29.960 --> 00:08:34.659 all the HTTP part or the network layer 00:08:35.300 --> 00:08:39.780 is running completely out of the interpreter, 00:08:40.940 --> 00:08:46.020 which kind of alleviates the load on the interpreter, 00:08:46.230 --> 00:08:47.320 like on the Python side. 00:08:48.550 --> 00:08:50.680 And depending on the context, 00:08:50.860 --> 00:08:53.420 it might be more performant 00:08:53.660 --> 00:08:57.500 or it might stabilize your latency more. 00:08:58.660 --> 00:09:03.360 You know, I think thinking back now, I'm pretty sure that that was the reason I chose Granian. 00:09:03.770 --> 00:09:09.420 Not that it was a ton faster, but the P95 bad side of things. 00:09:09.550 --> 00:09:13.260 Like how much might it slow down under certain weird circumstances? 00:09:13.760 --> 00:09:17.800 It was way more stable than a lot of the other servers, right? 00:09:17.870 --> 00:09:19.220 And that's kind of what you're referring to. 00:09:19.820 --> 00:09:23.640 Yeah, because again, like a normal server which runs everything into Python, 00:09:23.720 --> 00:09:39.480 Like it has this kind of side effects that when your application is starting becoming slow, it kind of becomes like an infinite loop in which like the server adds additional load to the interpreter, which is already like highly loaded from your application. 00:09:39.860 --> 00:09:49.940 And so like Graniac kind of avoids all of that because like even if your application is fully loaded, like all the web and network layer is out of the interpreter. 00:09:50.100 --> 00:09:52.880 It doesn't disturb, let's say the interpreter with that. 00:09:53.200 --> 00:09:53.280 Yeah. 00:09:53.540 --> 00:09:58.080 I do want to just give a little bit of a hat tip to WSGI or WSGI. 00:09:58.450 --> 00:10:14.240 Because, yeah, you pointed out it is old, but it's really, really nice that if somebody is out there running on GUnicorn or they're running on some other server, MicroWSGI, which you should stop running MicroWSGI because it's not supported anymore. 00:10:14.600 --> 00:10:16.580 You don't have to change your code at all. 00:10:16.760 --> 00:10:20.480 You just take your flask or your Jenga or whatever, and you just say, now run here. 00:10:20.510 --> 00:10:24.740 And they just, Grandin talks WizGey or Asggy to that thing, and it's transparent. 00:10:25.330 --> 00:10:25.720 It's beautiful. 00:10:26.140 --> 00:10:26.240 Yeah. 00:10:26.510 --> 00:10:26.620 Yeah. 00:10:26.940 --> 00:10:33.659 I mean, there are like still some, you know, sometimes weird things like, like, Vertzoic 00:10:34.500 --> 00:10:37.680 has some subtleties, sometimes some nuances. 00:10:37.840 --> 00:10:50.700 So I had to like add some environment key just for Verzoic because under some circumstances, if that key is not there, things don't work exactly good. 00:10:51.120 --> 00:11:02.120 So yes, I agree that like having like the two protocols separated and coded like in a specific way gives you this ability to switch the server. 00:11:02.520 --> 00:11:08.180 On the other hand, sometimes, you know, you can be a pain in some time. 00:11:08.680 --> 00:11:09.400 Yeah, for sure. 00:11:09.840 --> 00:11:10.460 Yeah, for sure. 00:11:11.780 --> 00:11:18.380 But yeah, I'd say like we like recently it's getting more popular. 00:11:18.690 --> 00:11:27.540 So I'd say if I'm not mistaken, like a few weeks ago, I think I surpassed like HyperCorn in terms of downloads. 00:11:28.760 --> 00:11:28.980 Awesome. 00:11:29.060 --> 00:11:37.480 And yeah, it's used like in a bunch of different big companies, Microsoft, Google, for sure. 00:11:38.350 --> 00:11:40.000 We use it at Sentry, of course. 00:11:41.140 --> 00:11:46.040 And I think like Sentry is the only company in which we use like all the three supported 00:11:46.380 --> 00:11:51.300 protocols of Granian because Granian also supports like its own protocol, which is named 00:11:51.410 --> 00:11:56.320 like RSGI, I don't know, like RSGI, which is a redesign. 00:11:56.740 --> 00:11:56.940 Risky. 00:11:57.840 --> 00:11:57.940 Yeah. 00:12:00.220 --> 00:12:07.460 um so yeah i'd say it's going pretty good um i think like the last let's say this year was mostly 00:12:07.800 --> 00:12:16.340 into um not not not not mostly about features i think features but mostly about you know making 00:12:16.480 --> 00:12:22.400 the thing more stable and covering like some edge cases um especially again we're using it the 00:12:22.460 --> 00:12:28.260 Sentry. So now like if something is wrong with Grinia, like I have people knocking on my door. 00:12:28.900 --> 00:12:34.900 I bet you do. Well, I mean, with Sentry, Sentry gets an insane amount of like an unimaginable 00:12:35.040 --> 00:12:39.480 amount of traffic. I'm sure with the error reporting side of things and probably a little 00:12:39.580 --> 00:12:46.040 more so these days on the MCP and API side as well with all the agents. I know that I use the MCP 00:12:46.120 --> 00:12:50.580 and absolutely love it. It's been incredible. It's really, really good. I could just say on my 00:12:50.600 --> 00:12:54.720 project hey there's a century error i got what's up like literally that's all i gotta say in clause 00:12:54.840 --> 00:12:57.920 like hold on we're on it and it's like yeah you know five minutes later it's like here's what's 00:12:57.930 --> 00:13:02.580 going on let's work on it you know yeah really good stuff uh i do want to give a quick shout out as 00:13:02.700 --> 00:13:08.220 well uh on this regard to gradient like people are like oh it's it's not as popular as server x or 00:13:08.310 --> 00:13:12.120 whatever right although i think that's starting to fade it's it's got 5 000 stars and a ton of 00:13:12.210 --> 00:13:19.079 people using it and so on yeah but it's also based on hyper from rust which itself has uh where where's 00:13:19.100 --> 00:13:25.420 my numbers, 410,000 projects using it. So it's really tested, right? And obviously the stuff at 00:13:25.500 --> 00:13:31.640 Sentry as well. Yeah. Yeah. Like the only thing I'm sure about Granian is that like the HTTP stack, 00:13:32.100 --> 00:13:38.280 that's super safe. Like that's like, I could bet like everything I have on that. So if something 00:13:38.300 --> 00:13:44.660 is wrong in Granian, like that's on me, not on Hyper, like to be clear. This portion of Talk 00:13:44.660 --> 00:13:49.660 Python to me is brought to you by Sentry and Sear AI. There are plenty of AI tools that help you 00:13:50.080 --> 00:13:55.280 write code, but Sentry Sear is built to help you fix it when it breaks. The difference is context. 00:13:55.920 --> 00:14:00.500 Sear isn't just guessing based on syntax. It's analyzing your actual Sentry data, 00:14:00.760 --> 00:14:05.360 your stack traces, logs, and failure patterns. Because it has the full context, it can 00:14:05.740 --> 00:14:09.620 A, spot buggy code in review and help prevent issues before they happen, 00:14:10.260 --> 00:14:13.040 and B, identify the root cause of production errors. 00:14:13.980 --> 00:14:18.540 It can even draft a fix and hand the work off to an agent-like cursor to open a PR for you. 00:14:19.140 --> 00:14:21.140 Seer turns Sentry into a complete loop. 00:14:21.660 --> 00:14:24.500 You have your traces, errors, logs, and replays to see the problem, 00:14:24.800 --> 00:14:26.260 and now AI to help solve it. 00:14:26.620 --> 00:14:30.920 Join millions of devs at companies like Claude, Disney+, and even Talk Python, 00:14:31.380 --> 00:14:32.560 who use Sentry to move faster. 00:14:33.080 --> 00:14:37.860 Check them out at talkpython.fm/sentry and use code talkpython26, 00:14:38.440 --> 00:14:41.800 all one word, for $100 in Sentry credits. 00:14:42.440 --> 00:14:44.180 Thank you to Sentry for supporting Talk Python. 00:14:45.660 --> 00:14:48.980 I recently did a post over on my personal blog 00:14:49.080 --> 00:14:52.860 calling cutting Python web app memory by 31%. 00:14:52.860 --> 00:14:55.760 And I did a ton of analysis and stuff 00:14:56.660 --> 00:14:59.260 on how I was running Granian and different things, 00:14:59.300 --> 00:15:00.240 just how I was running Python. 00:15:00.820 --> 00:15:02.400 And I think this is going to be a good lead-in 00:15:02.400 --> 00:15:03.540 to where we're going as well. 00:15:03.880 --> 00:15:13.040 I got it to where I was using 1.3 gigs, let's say, for Talk Python training the courses specifically. 00:15:13.360 --> 00:15:15.160 And people are like, oh, it's just a little courses thing. 00:15:15.200 --> 00:15:19.960 Like, well, it has 178,000 lines of Python, so it's not a totally small project. 00:15:20.440 --> 00:15:23.240 But down by the end, I did a bunch of stuff on it. 00:15:24.480 --> 00:15:25.280 Lazy imports. 00:15:25.779 --> 00:15:28.660 I switched to async, which is why I brought this up. 00:15:28.940 --> 00:15:31.160 But somewhere down here, I don't know what the final number is. 00:15:31.400 --> 00:15:31.820 Here we go. 00:15:31.880 --> 00:15:36.560 In the end, I got it down to 450 megs from 0.28 gigs. 00:15:37.190 --> 00:15:40.900 One of the main reasons I did that as a way I want to kind of want to lead into like this, 00:15:41.380 --> 00:15:45.740 where we're going is I said, well, the way grading works is for each, 00:15:46.250 --> 00:15:49.540 either you can make multi-processes that are single threaded or it could do multiple threads 00:15:49.660 --> 00:15:51.380 and then it'll create a separate interpreter for each one. 00:15:51.490 --> 00:15:55.240 Either way, it kind of like replicates out the running. 00:15:55.420 --> 00:15:58.660 It'll create copies in one way or another of your code. 00:15:59.010 --> 00:16:01.760 And that, if you could run one of them, that would be great. 00:16:01.840 --> 00:16:03.000 But why do we have so many? 00:16:03.330 --> 00:16:04.900 We have so many because of the GIL. 00:16:05.010 --> 00:16:06.140 We have poor concurrency. 00:16:06.640 --> 00:16:10.360 It's really hard to handle a lot of load and like actually access, you know, my servers, 00:16:10.600 --> 00:16:10.920 eight cores. 00:16:11.010 --> 00:16:14.060 Like I could only access one core like normally, right? 00:16:14.380 --> 00:16:17.560 And so we do all these things to like explode them out. 00:16:17.680 --> 00:16:18.880 But what is the most expensive? 00:16:19.140 --> 00:16:22.160 Like five years ago, this was true on servers, but really is true now. 00:16:22.250 --> 00:16:23.000 What is most expensive? 00:16:23.360 --> 00:16:24.380 Memory, not compute, right? 00:16:25.180 --> 00:16:25.300 Yeah. 00:16:25.700 --> 00:16:25.760 Yeah. 00:16:26.160 --> 00:16:26.600 Thanks AI. 00:16:27.000 --> 00:16:27.540 Thanks AI. 00:16:27.710 --> 00:16:27.940 Thanks Altman. 00:16:28.860 --> 00:16:30.120 We really appreciate that. 00:16:32.260 --> 00:16:34.580 But we've free-threaded Python now. 00:16:34.780 --> 00:16:35.300 That's a thing. 00:16:35.740 --> 00:16:43.280 And so we're kind of moving into this world where we can take better advantage of the hardware and we don't have to multiply out. 00:16:44.220 --> 00:16:47.060 I know your process takes 500 megs of RAM, but we need four of them. 00:16:47.090 --> 00:16:49.580 So now you need two gigs just to run your code. 00:16:50.360 --> 00:16:51.940 We don't necessarily need that anymore. 00:16:52.160 --> 00:16:53.780 And grading would be a really good foundation for that. 00:16:54.000 --> 00:16:59.440 But let's just think broadly about free-threading and async as it is and stuff. 00:16:59.660 --> 00:17:02.620 I know you've thought a lot about this because of this project that you built. 00:17:03.680 --> 00:17:04.819 Well, what do you think? 00:17:05.350 --> 00:17:08.720 First thing, can I say free threading to me? 00:17:09.140 --> 00:17:11.199 It's super exciting. 00:17:11.640 --> 00:17:16.060 And I don't know why people are not as much excited as me on this. 00:17:16.819 --> 00:17:25.480 Because this is, I'd say, the biggest thing that ever happened into Python since the beginning of Python. 00:17:25.680 --> 00:17:27.140 In the last 30 years. 00:17:27.380 --> 00:17:31.720 do you remember like Python 2 to Python 3 and all like, 00:17:32.080 --> 00:17:36.360 oh, now strings are not bytes anymore and everything is Unicode? 00:17:36.700 --> 00:17:36.800 Yeah. 00:17:37.040 --> 00:17:40.960 That's nothing compared to, hey, the gil is not there anymore. 00:17:41.080 --> 00:17:46.800 Like we talk about like every Python room everywhere in the world, 00:17:46.860 --> 00:17:53.220 like at some point talked about the gil and why we have to do to have the gil, right? 00:17:53.960 --> 00:17:54.160 Right. 00:17:54.720 --> 00:17:59.220 The Gilectomy that Larry Hastings had, it seemed to perpetually be working on and never 00:17:59.440 --> 00:18:00.020 kind of went anywhere. 00:18:00.340 --> 00:18:00.480 Yeah. 00:18:00.480 --> 00:18:05.660 It took like 15 years of people talking about this and it finally landed. 00:18:05.860 --> 00:18:12.000 Like last year, yes, of course, last year it was beta because like in Python 3.13, it 00:18:12.000 --> 00:18:14.380 was like, this is an experiment still. 00:18:14.900 --> 00:18:17.840 That was the asterisk on top. 00:18:18.960 --> 00:18:22.120 But since Python 3.14, that's considered stable. 00:18:22.540 --> 00:18:28.560 It's exactly as the normal GIL-included interpreter. 00:18:30.480 --> 00:18:33.540 And at the end of the day, the biggest part of all of this 00:18:33.800 --> 00:18:38.800 was using reference counting instead of singleton. 00:18:39.160 --> 00:18:44.000 So I guess I'm excited about free-threaded Python 00:18:44.020 --> 00:18:47.840 because finally I have the language I love the most. 00:18:49.020 --> 00:18:53.180 not because of the structure, not because of types or how it's designed, 00:18:53.180 --> 00:18:57.840 but because of this awesome community that only Python language has, 00:18:58.220 --> 00:18:59.080 at least in my experience. 00:19:00.220 --> 00:19:04.080 And so finally, I can use this language with actual threads. 00:19:04.540 --> 00:19:08.460 So I'm not obliged anymore to think, oh, wait, we are in Python, 00:19:08.920 --> 00:19:13.760 so I need to use a C library or something to take advantage of my CPU, 00:19:14.480 --> 00:19:25.040 Especially given, I don't know, we are in the era of, hey, even if you buy a $200 CPU that probably has 12 cores or something. 00:19:26.380 --> 00:19:28.420 So yeah, I'm very excited. 00:19:29.700 --> 00:19:32.740 I'm confused about the lack of adoption. 00:19:33.160 --> 00:19:35.260 I would love to see more adoption in this. 00:19:35.500 --> 00:19:46.120 And of course, like there are reasons, like every native C compiled extension has to be fixed to be thread safe. 00:19:47.640 --> 00:19:53.360 And probably like on a bunch of ancient extensions, that's not simple or easy. 00:19:55.180 --> 00:19:56.240 Yeah, I don't know, man. 00:19:56.320 --> 00:19:58.100 Like I'm just too excited. 00:19:58.380 --> 00:20:00.620 Like I want people to use free thread. 00:20:00.800 --> 00:20:02.020 I know. 00:20:02.180 --> 00:20:04.260 And getting it is so much easier as well. 00:20:04.340 --> 00:20:10.840 it's just uv Python install, you know, that 3.14 T and you got it, right? 00:20:10.980 --> 00:20:11.200 Yeah. 00:20:11.600 --> 00:20:11.720 Yeah. 00:20:12.080 --> 00:20:16.740 And like, I don't know, I'm so pissed at the maintainer of the Docker images 00:20:16.880 --> 00:20:20.240 because we don't have like an official 3.14 free threaded image. 00:20:20.360 --> 00:20:20.660 Like why? 00:20:21.160 --> 00:20:21.220 Why? 00:20:21.400 --> 00:20:23.460 Oh yeah, that is actually a good point. 00:20:23.580 --> 00:20:28.900 Like it has been a year and a half and it just requires putting the T on the uv installed bit. 00:20:29.160 --> 00:20:29.440 Yeah. 00:20:29.740 --> 00:20:29.860 Yeah. 00:20:30.320 --> 00:20:30.460 Yeah. 00:20:30.480 --> 00:20:36.660 Yeah, but again, to my perspective, free-threaded Python opens the window to do a bunch of... 00:20:36.840 --> 00:20:42.040 First, to solve a bunch of the issues you mentioned, which is, for example, in Centri, 00:20:42.160 --> 00:20:43.980 the number one issue is... 00:20:44.120 --> 00:20:50.460 I mean, Centri at the end of the day is this 18-years-old Django monolith. 00:20:51.420 --> 00:20:59.520 And so, yes, we deploy in several multiple different deployments, activating just some parts. 00:20:59.720 --> 00:21:20.080 But even so, like, yeah, we have the same problem, which is, hey, like, in order to use like one C view out of like a VM that GCP gives us, like, yeah, we need to spawn like four different processes and require like, I don't know, four, five, six gigabytes of memory for like a single pod. 00:21:20.320 --> 00:21:31.320 And not only do you have to do that, you also have to, anytime there's some sort of in-memory cache, like an LRU cache or something, every one of those has to be spun up for all four, not just for one, right? There's a lot of powerful sharing. 00:21:31.440 --> 00:21:34.120 Do we want to start talking about observability? 00:21:34.420 --> 00:21:36.580 Because you cannot just use like 00:21:36.640 --> 00:21:39.380 the Prometheus SDK or anything like that. 00:21:40.000 --> 00:21:42.000 Because again, like then those became like 00:21:42.480 --> 00:21:43.340 four different instances. 00:21:43.600 --> 00:21:45.800 So you cannot even track like the same metrics 00:21:46.620 --> 00:21:48.480 unless you do like some weird tricks. 00:21:48.960 --> 00:21:53.200 Like, yeah, relying on the multiprocessing library. 00:21:54.180 --> 00:21:54.800 It's messy, man. 00:21:54.980 --> 00:21:57.080 So like, yeah, I think like 00:21:57.620 --> 00:22:00.220 FreeThreader like solves a bunch of problems, 00:22:00.820 --> 00:22:04.180 Especially in web development in Python. 00:22:05.420 --> 00:22:09.780 Because everybody says like, oh yeah, they did this just because of AI, you know, 00:22:09.880 --> 00:22:12.240 because they can like make things parallel, whatever. 00:22:12.720 --> 00:22:14.200 I don't get that. 00:22:14.300 --> 00:22:18.280 Like, like we have like this amazing feature and it solves a bunch of problem 00:22:18.460 --> 00:22:23.180 on web development and also like it opens up the a bunch of new opportunities. 00:22:23.400 --> 00:22:27.680 Because again, like one of the most annoying parts in Granian, which is also 00:22:27.700 --> 00:22:33.400 Like one of the things that people opens like the most, the vast majority of issues about 00:22:33.580 --> 00:22:38.960 is because nobody understands like how socket sharing works across different processes. 00:22:39.370 --> 00:22:45.100 And again, like with FreeThreader, like that's just go away if we have something different 00:22:45.230 --> 00:22:45.940 from a Synco. 00:22:46.310 --> 00:22:51.280 And I guess that's the, that's when we get on the point here. 00:22:52.520 --> 00:22:57.660 Because for example, like in Grainium, so Grainium supports FreeThreader Python scenes 00:22:57.680 --> 00:22:59.440 2.0, if I'm not mistaken. 00:22:59.920 --> 00:23:01.520 And so when you use 00:23:01.700 --> 00:23:03.420 Granian on a free-threaded interpreter, 00:23:03.840 --> 00:23:05.520 workers become 00:23:07.600 --> 00:23:07.660 threads. 00:23:08.200 --> 00:23:09.340 They're no longer processes. 00:23:10.200 --> 00:23:11.520 And I can skip all the dance 00:23:11.620 --> 00:23:13.160 of, yeah, open the socket, 00:23:14.140 --> 00:23:15.500 but just bind 00:23:15.500 --> 00:23:17.440 on that, but not on listen, because then 00:23:17.640 --> 00:23:19.440 it gets to be shared across processes 00:23:19.960 --> 00:23:21.500 and so each worker can have his own 00:23:21.660 --> 00:23:23.540 backlog. It's a mess, man. 00:23:23.740 --> 00:23:25.700 So on the free-threaded variant, 00:23:26.200 --> 00:23:28.100 Like, Grenian already, like, used threads. 00:23:29.799 --> 00:23:35.280 The subtlety is that each thread has its own event loop. 00:23:35.510 --> 00:23:41.280 So what Grenian cannot do today is sharing things between those threads. 00:23:41.560 --> 00:23:49.180 At the end of the day, yes, the application is shared, but, like, every request, like, if a repus comes to a worker, that's it. 00:23:49.250 --> 00:23:50.120 Like, there's no... 00:23:50.310 --> 00:23:51.780 You cannot cross boundaries, right? 00:23:52.060 --> 00:23:54.040 when you await for something, 00:23:54.130 --> 00:23:55.260 when you suspend for something, 00:23:56.440 --> 00:23:57.500 that's the loop, 00:23:57.750 --> 00:23:59.020 the event loop you're running into. 00:23:59.980 --> 00:24:01.540 Which is a big limitation for servers. 00:24:02.300 --> 00:24:03.320 Yeah, and one is calling await 00:24:03.540 --> 00:24:04.400 and another is calling await 00:24:04.680 --> 00:24:06.060 or maybe doing some work, 00:24:06.260 --> 00:24:07.860 but if they were the same event loop, 00:24:08.220 --> 00:24:09.180 one thing could run 00:24:09.420 --> 00:24:10.240 because the other is awaiting, 00:24:10.380 --> 00:24:12.400 but they're actually separate altogether, right? 00:24:12.560 --> 00:24:15.460 Yeah, and also you can still end up 00:24:15.460 --> 00:24:16.720 in these weird conditions 00:24:16.930 --> 00:24:19.360 in which maybe, I don't know, 00:24:19.440 --> 00:24:21.139 50% of your connection 00:24:21.740 --> 00:24:24.640 we're bound to a specific event loop. 00:24:25.180 --> 00:24:28.080 And so you also have like unbalancement 00:24:28.190 --> 00:24:29.420 between like the different threads. 00:24:29.510 --> 00:24:30.640 So you have like one thread 00:24:30.800 --> 00:24:32.880 that receives the vast majority of things, 00:24:33.060 --> 00:24:34.100 the other thread doing nothing. 00:24:35.480 --> 00:24:35.640 Yeah. 00:24:35.860 --> 00:24:36.000 Yeah. 00:24:36.390 --> 00:24:38.380 Let's say like we have new opportunities. 00:24:39.200 --> 00:24:41.080 We kind of lack some libraries, 00:24:41.370 --> 00:24:42.700 like some libraries behavior 00:24:43.700 --> 00:24:45.620 that can actually give us like 00:24:45.880 --> 00:24:47.580 all the actual meat here. 00:24:48.600 --> 00:24:48.900 Agreed. 00:24:49.020 --> 00:24:53.600 I want to talk about one area before we dive in, the Tone.io, 00:24:53.820 --> 00:24:55.740 and that's just asyncio event loops. 00:24:56.120 --> 00:24:57.460 And I know during your presentation, 00:24:57.700 --> 00:25:00.240